Adaptive Learning in Nonlinear Models

Two separate branches of research have shown the empirical benefits of nonlinearities in macroeconomic models and bounded rationality, respectively. This paper bridges these literatures by developing a methodology for estimating DSGE models when expectations are formed by adaptive learning within a nonlinear approximation of a model. This novel approach is applied to a standard New Keynesian model that is estimated using U.S. data. The nonlinear model with adaptive learning improves model fit compared to the linearized model with and without adaptive learning as well as the nonlinear model with rational expectations. These results are stronger when nonlinear features are added to the model and when including data where interest rates reach the zero lower bound